Sorry, I don't buy this time artifacts are relevant at all:

1.- they happen on very rare occasions with real music (only if there are high levels and of impulsive type at a very specific frequency around 22 KHz)

2.- they happen also at the AD stage, because the signal is first brickwall-filtered at this stage. At the DA stage, the ringing may not occur at all, depending on how the filtering was done at the AD stage (due to possible filtering at the AD stage of the precise frequency that "excites" the ringing at the DA stage... but also possible that the ringing from both stages adds up, I have to investigate more on this (*1).). Any ringing may appear is very likely to be already recorded on the cd, and cannot be avoided just by not filtering at his stage.

3.-when they happen, are not audible at all due to (*2) its short duration and backwards temporal masking.


So there's a clear pick in my opinion.



(*1) Edit: I've done a quick test with CEP filters, and seems that the ringing doesn't add up.

(*2) Edit: Also, of course, due to the frequency of the pre-ringing (~22 KHz) which falls out of the audible band.

Edit: this nasty quotes break whenever you edit the message (Garf: this better? KikeG: Sure )

The guy from TNT audio interviewing this japanese says it, and you should listen closely :

That is not said at all !!

There haven been a number of publications in the past 10 years about the capability of the Human ear to 'hear' sound > 20 KHz ( dont ask me for any links, i dont have any avaialble )

The trick is to differentiate between

- static ( sine )

and

- dynamic ( transient, impulses )

'hearing' .

I did tests with my own ears, almost 5 years ago :

- I was not able to 'hear' sine tones > 16.5 KHz ( maybe even lower now )

but

- in a blind ABX test i could easily detect a low pass filter of 4th order ( 24 db/octave , linkwitz ) with a -3 dB at 20 KHz on a high quality Stereo chain, with normal music ( hitrate : > 85 % ) .

Why ?

There are theories taking into account that we dont have only one ear, but two of them. Human brains can localize sound sources very precisely, especially if the sound is transient, means no sine tones. Dont ask me how its done exactly, if i remember correctly its a kind of measuring the time difference between the signal arriving on the one ear and the other time short after ( a few µs resolution ).

This theory, if i remember correctly, now states that this time resolution works better the higher the frequency of the transient is. A common example will be that its easier to localize the sound produced by a metal hammer hitting metal, compared to a soft drumstick hitting a drum.

Means you cant really her 'sound' or 'tones' > 20 KHz, but the missing/existence of sound > 20 Khz can deteriorate sound localisation. BTW, its not said that sound > 20 Khz cant be detected at all because of the inner ear. We have a small membrane in our inner ear ( dont ask me for its name ) being positioned before cochlea, and its function is still not 100% clear to science.

I would also think that you're most likely to hear ultrasound during transients if you can hear ultrasound at all.

I only know of one study done by japanese researchers, where they "measured" different brain waves when the test subject was exposed to high ultrasonic levels. However, the subject didnt hear anything, and the different brain waves were present even after the utrasonic sound ceased. No one was able to repeat this experiment, iirc.

If you know of another REAL study about this, it would be interesting to know about it. But not of the type "I've heard" or "I've read somewhere" myths.



Maybe your filter with -3 dB at 20 KHz was something like -1 dB at 13 KHz, I don't know. This would be quite audible with harmonic-type musical signals.

The thing is that I can hear up to 18 - 18.5 KHz with sine signals, but can't differentiate an impulsive clip such as castanets attack brickwall-filter lowpassed at 15 KHz from non-lowpassed. Even have some difficulties with a 12 KHz lowpass. Just try at http://www.pcabx.com/technical/low_pass/index.htm. Just try to ABX any of the 18 KHz lowpassed clips, and tell me if you are able to hear them sound different. Go down into the lowpass until you can hear the difference.

This is because with this more complex real world signals, masking comes into effect, and the lower high frequencies easily mask the higher high frequencies. This doesn't happen with pure tones (sines).

About localization, what happens is that ear is very sensitive to interaural delays, but if you lowpass both channels the same way, the interaural delays remain the same. Even when ear is so sensitive to interaural delays, this doesn't has much to do with the upper frequency limit of hearing.

What kind of filter was it ? Analog or digital ?

How much ABX trials were there ?

it was an analog filter i had done using a few op-amps. The interesting thing was that the filter was always in the signal path, but only cut-off frequency was changed from 20 to 40 Khz with a normal switch.

I made my brother switch to any state without telling me, and did 3 runs alltogether, each being 20 repetitions.

A 24 db linkwith filter shouldnt have a 1 db attenuation at 13 KHz.

About the ABX test : i dont have any suitable stereo system connected to my PC for the time being ( and no plans to do so ), so i cant participate in any tests right now. I just moved to another house recently, with a much bigger living room, and my 'good' stereo system will hopefully soon have a 2nd spring. Now the only question is how i could connect a decent 24/96 soundcard ( which i also dont have for the time being ) to this system ?

To make this clear ( i maybe wasnt before ) :

The comparison test mentioned above was done with analog material coming from a Linn LP 12, Karma pickup, amplified with Muscial Fidelity Preamp, and playing direct cut material.

Of course, this test was useless with CD material, as it doesnt contain anything > 20 Khz . The only sensible way to make such a test on a PC would be with a decent 24/96 soundcard and music recorded at this very sampling rate.

You cant hear what i decribed with test signals ! The recording was also very important IMHO, as most studios destroy the original stereo information by recording every instrument with a dedicated microfone and giving its 'position' in the stereo image using the 'balance' slider .... so there is no runtime difference between left and right ear, but simply an amplitude difference ...

Sheesh... all of this started from an original question, "is SACD good or bad?" (LOL). Well, why not just arrange to listen to one (with a good recording and proper equipment) and decide that way? End of thread.

You can't say "most" studios use this. Or just give an example (well, ok, this is easy since fewtch just posted one yesterday )
Among my 200 CDs of mostly electronic or pop music, I think I can find two or three recorded in mono, one or two in pure stereo, and maybe one or two with a "balance slider stereo" as you describe.
Most of them use an artificial DSP stereo with delay, spatialization, reverb and all.

bleep

You're right, I didn't realize it was a 24 dB/octave filter when I replied.

However, other things could have affected your test. Non perfect filter response, for example (higher or lower gain on the passband depending on the cutoff, big phase aberrations, etc.), because I don't believe a 20 KHz lowpass can be easily audible at all.

Try when you can the test I propose, the results are quite convincing. If you can't hear a 19 KHz tone, and you can't hear a 16 KHz lowpass on a signal with many transients (castanets), it does not seem logic that you could hear a 20 KHz lowpass on that same signal.

You can try on more critical signals, possibly such as some cymbals (maybe castanets has too much low frequency impulsive content) , I believe Pio has some available, lowpass them with something like 18 KHz an see if you can tell the difference.

Also, 85% hitrate doesn't mean much by itself, it would be necessary to know the number of trials to see how significan is that number.

In the same website (pcabx) there was a test similar to yours, but comparing 24/96 musical signals with 16/44.1 signals, unfortunately, it seems to be offline now. As I know, nobody had been able to identify those different formats in an ABX test.

What kind of 'test signals' were used here ?

If it was a kind of single instrument i dont think you could 'hear' what i could differentiate in my comparison, being the 'air' around instruments, the precision of positioning and the 'depth' of the recording studio .

Chesky recordings, again, are very accurate in stereo positioning and delays ( maybe they use a good old stereo microphone assembly, like a Jecklin configuration, at least for some recordings like Ana Caram )

He said analog filter. The most likely explanation for him ABXing the lowpass is phase distortion. QED

The 'air' around instruments has not much to do with the upper high frequency range, is usually more a thing of middle and upper-middle frequencies.

Try dowloading the "rach_original.flac" file from http://www.ff123.net/samples.html

It is a quality, 'airy' piece of classical music from Telarc records. If you analyze it, you'll see that it has practically no content above 10 KHz.

About your test, vinyl can go up to 50 KHz in some cases, but there is just garbage at there. I beleiev max. "usable" frequency is below max. usable frequency at cd (which goes up to near 22 KHz).

If it was so easy to detect a 20 KHz lowpass, why is it so universally agreed between audio engineers and researchers that our hearing only goes up to that frequency, in best cases? Why are there no experiments *clearly* proving that we can "feel" (=hear) things above 20 KHz?

Back online :



Right-click and save target as cosmic.pac (524 ko)



Right-click and save target as transwave.pac (991 ko)

Air liquide.pac was impressive in the 14 kHz range, it must mask higher frequencies. I didn't bring it back.

Nobody tried these samples yet. I don't know if they are good test samples for lowpass, but at least they seem to.

KikeG,

my knowledge about audio engineering can certainly not compare to yours, but if you are telling me that it is impossible for humans to 'take audible information' ( maybe better then using the expression 'to hear' ) from the frequency range > 20 Khz by looking at a track with a spectrum analyzer, than it seems i failed to communicate the idea behind my original statement, being why frequency components in 'sound transients' maybe could be helpful for human hearing 'system' ( thats is including brains ) to support sound localisation, and thus to improve 'spatial' impression of music ....

Well, I was just intending to explain that this "airy" sensation between musical instruments, as far as I know, has not much to do with the upper frequency range of hearing.

About looking a track with a spectrum analyzer... well, sound at last is only a signal, some of whose properties can be easily analyzed. If a classical musical piece has no content over 10 KHz, you can bet it won't have over 20 KHz.

About frecuencies over 20 KHz... well, if we are *totally* deaf, unsensitive, to a ultra-high level signal of 21 KHz, I can't see how, much lower levels that can be found in regular music at these frequencies, could have an effect over our perception.

Regarding the airy sensation, I concur with KikeG.
I've also created it while trying ways to add ambience to a dry mono vocal recording at 11.025 kHz sampling rate, 16-bit/channel without comb-filtering or colouration of the sound.

This sampling rate allows only frequencies of 5.5 kHz and below, and I've resampled properly to 44.1 kHz (bandwidth limited) after adding the effect to guarantee no aliasing, and the effect is still there (not an artifact of my soundcard operating at low sample rates where it's not anti-aliased.

It's a kind of airy quality - a gentle whisper and feeling of space around the sound. It feels slightly grainy and analogue, and much more like being in an auditorium - I guess that's a sort of "presence".

That wasn't audible in the original sound, yet all frequencies above 5.5 kHz have been filtered out, so despite the hissy nature of the "grain" or "air" that suggests a wide frequency response, it's not actually a wide response at all, it's somehow in the ambient relfections I aritificially added and can be represented within the frequencies below 5.5 kHz.

Incidentally, I have some other ideas from this exercise - including the use of complex numbers (Argand diagrams) to represent stereo waveforms, and how this might allow an alternative, safer method of creating monaural recordings from stereo sound. If I get time to work this through, I'll share it.

In the mean time, I'll open up a new thread to describe the ambience effect.

Dick Darlington

Here is my little theory about this. The human animal evolved from others animal that used to live in water. And in water the sound you can "hear" is much higher in the frequency range. The same way as our brain have a reptilian part (this is the technical name), maybe our body have some reminiscents of our origins.

Now about how we could hear transients. If you take a constant sound at 15kHz that you can hear. The internal parts of your hear will move with approximately the same movement/speed. We basically can't "hear" much higher because of the mechanical inertia of the moving parts in our ears... But if you have a short click (a few milliseconds) at 30 kHz. There will be the moment, when the sound starts, that your ear will move. But it won't be able to follow the speed. That doesn't mean your ear will not move. Especially if the frequency is not 100% constant (it will produce modulation at low frequencies). The movement of the ear, might not be at all like the movement of the air we "receive", but in the end there is one. And I think we can feel it.

Maybe that's all bullshit, but I think it's unfair to say that we cannot feel at much higher frequencies. I believe Christian was fair with his tests and that it somehow proves that the frequency range is more complex that what we currently know. (and I don't say that because Christian is my working partner )

Yes, my prof in uni always was trying to make clear to us that there is a HUGE difference between the hadnling of the ear for static ( sine ) and dynamic ( transients ) signals.

BTW, high frequency hearing tests with earphones will fail completely to be able to find the effects i am proposing, as most test signals are mono, so there is no form of time delay between L and R ....

Most of the ideas that are being kicked around here could be proved or disproved in a well planned experiment.

And some people here have the equipment to try such an experiment.

1. 2496 sound card
2. decent headphones
3. Cool Edit


We've been discussing these things for ages. When these ideas were first kicked around in the 1990s, no "normal" people had the equipment to be able to test the ideas. Now, many people do. It seems to me, rather than discussing hypotheticals, we could also come up with some ways of proving/disproving the hypothesis, and then carrying out the experiment.


As my own contribution, the 1st hypothesis seems to be:

24/96 material sounds better than 44.1/16

The approach should be
a) get 24/96 material (David Chesky sells some nice DVDs)
B) convert to 44.1/16 (Cool Edit will happily do this as well as anything)
c) compare (blind test)

You'll have to remove all confounding problems - for example, I can hear when my audiophile24/96 switches sample rate, which makes a blind ABX test rather difficult! Putting some silence at the start of tracks (and not switching mid-track) may remove this problem.


If, and only if, someone can hear a difference, then we can start trying to figure out what mechanism is causing that difference.

Any takers?

Cheers,
David.
P.S. - When I had decent headphones, I didn't have the 24/96 sound card - now I have the card, I no longer have the headphones. However, I'm working on it! I've never had a quiet listening environment.

http://www.hfm-detmold.de/texts/de/hfm/eti...ten/seite2.html


"..............dass durchschnittlich Frequenzen ab 16-17 kHz nicht mehr wahrgenommen werden. In seltenen Fällen können Personen 20 kHz und darüber detektieren. Diese Angaben beziehen sich aber nur auf die Eigenschaften des Gehörs bei periodischen Stimuli. Da Musik und die meisten Schallereignisse aber nicht nur aus periodischen Vorgängen bestehen, sondern durch unperiodische Einschwingvorgänge charakterisiert werden, muss das Gehör auch auf die Aufnahme und Verarbeitung solcher unperiodischer Signale hin überprüft werden.

Untersuchungen, die speziell mit unperiodischen Signalen und steilen Impulsen als Reiz arbeiten, sind vergleichsweise jung und werfen noch viele Fragen auf. Impulshafte Signale, wie etwa ein Trommelschlag, können in ihrem scharfen Einschwingvorgang sehr hohe Frequenzen über 20 kHz im Frequenzbereich enthalten. Es ist also durchaus denkbar, dass während geräuschhafter Einschwingvorgänge der auditiven Wahrnehmung die Möglichkeit gegeben ist, indirekt auch Frequenzen über 20 kHz zu verarbeiten und zu einer bewussten klanglichen Unterscheidung zu führen.........."


Short translation of this part of a Diploma work:

average listeners may listen up to 16-17 kHz, not more !
Some persons up to 20kHz and a little bit more.
This refers to abilities of "ears/brain" regarding periodic stimulation.
As music does not always consist of periodic stimulation, but contains aperiodic "Einschwingvorgänge = first-time-oscillation", the hearing has to be tested to aperiodic signals.

Tests, surveys regarding aperiodics, are young and result to new questions.
Impulse signals like drum, have a sharp "first-time-oscillation" (aperiodic) , and may contain much higher frequencies than 20 kHz.
It may be possible, that "listening" of frequencies higher than 20 kHz may be possible, during these "aperiodic first-time-oscillations".

So, they tested the high-sampling frequency of 96 kHz vs. 48 kHZ with practical music, with the result, that 48 khZ sampling is enough, indishtinguable from original sound.
Test-listeners were a good group of students becoming recording-ingeneers.

They were using a tweeter fast enough to reproduce transients containing high frequencies, like a Manger or an airmotion transformer, in the listening tests ?

Note that a normal studio monitor is just not good enough for these kind of experiments .... good earphones, like some nice STAXes, should be fine .....

Wrong.

You must compare 96/24 and 96/24->44.1/16->96/24 audio.

96 kHz and 44.1 kHz often sounds different (with problems on the 96 kHz
side), because the frequency response is very different.

1st series of tests: Compare 96/24 with 96/23...96/10.
2nd series of tests: Compare 96/24 with 96/24->64/24->96/24, 96/24->48/24->96/24, 96/24->44.1/24->96/24, 96/24->40/24->96/24, 96/24->32/24->96/24.

That would be a very interresting stuff. How many people here have the necessary hardware and willing to make that test ?

I've got a 24/96 soundcard, and HD 600 Headphones.

30 kHz at -3 db
38 kHz at -10 db

But I've got no 96 kHz material.

Robux4, what you describe about the transient response is perfectly right. But it doesn't support your point. A delayed slow response to a 30 kHz transient is nothing else than a clean lowpassed version of the transient. The pulse has a flat frequency content from 0 to 30 kHz. The delayed version has a flat frequency content from 0 to 16 kHz only. The removal of the inaudible frequencies acts exactly like inertia.
See a more detailed explanation at http://forum.cdfreaks.com/showthread.php?s...1864#post376461

However, the "periodic" behaviour against "transient" one is a valid question. In mathematical terms, we ask if the ear is a "linear" device. Linear meaning roughly that the effect of the sum equals the sum of the effects, and therefore a pulse being a sum of frequencies from 0 to N, if the ear is linear, it should react as a lowpass filter, the resulting lowpassed pulse actually heard being the "sum of the effects", that is, the sum of all reactions to audible content in the pulse.
But if the ear is not linear, it won't necessarily lowpass the pulse. The pulse heard won't necessarily be the sum of the effects that each frequency component in the pulse would have.

The most basic test is the intermodulation one. Say that you can't hear above 16 kHz. Generate a 6 kHz square wave in cooledit. And for god's sake, perform a spectrum analysis of it... I tried 3 times to generate square waves in Cool Edit before actually getting a true one. A true square of frequency N must have a frequency content of N, 3N, 5N, 7N ect... and nothing else ! The square generator in CoolEdit is completely wrecked. But it works at 6 kHz for a 48 kHz sampling rate.

Then, you have a 6 kHz sine, plus a 18 kHz one. If you compare it with a 6 kHz sine alone, all you'll hear is a volume difference (the square being louder because the level of the N frequency is above the peak level of the square).
Rising the volume, a ghost 12 kHz tone should appear. It's intermodulaton distortion in the ampli or speakers.
Now, playing a 6 kHz sine in one speaker and a 18 kHz one in the other, there is no intermodulation, no 12 kHz tone in the room.
WARNING : it is said that 18 kHz sines can fry tweeters !!!
Check with a microphone, that your speaker actually plays a 18 kHz tone.
If you hear a 12 kHz tone anyway, check with a microphone, that it is not in the room, and if it's not, you're hearing the effect of inaudible frequencies.
I've tried it, and could hear nothing at a level that would have been 100 db if it has been a CD playback. So tell me about 40 kHz @40 db !
I can't find the report of the other people that have tried the same experiment and failed too, but Nika says it's in there :
George, Watch this!!!....(96k) @George Massenburg, abstract in page 33



And last, there is Oohashi's experiment ( http://jn.physiology.org/cgi/content/full/83/6/3548 ) in which people say they couldn't hear the difference, but the electroencephalogram showed one (delayed). They say also that people said they were more comfortable when the ultrasonic content was played, but they don't say how many people said it. This should be analysed like an ABX result : if you ask people to choose which performance they prefer, and that a little more than half the people tell they prefer the ultrasonic one, it can very well be pure chance. They don't state the level of confidence of the correlation between people's statement and the performance played (=same kind as the level of confidence between your ABX choice and the samples really played).

I've just thought of a non linear behaviour of the ear... The ear, as a lowpass filter, should act like an IIR filter isn't it ? But a true linear behaviour would be the one of an FIR (well, a symmetric IIR one, with pre echo).
Wouldn't there be something here to discuss ?

IIR/FIR filters explanations :
The audio engineer's approach to understanding digital filters. For the idiot such as myself by Nika Aldrich

I agree pretty much with Pio's explanations and Frank proposals.

However, about the first, I think you don't have to confuse linear response with impulse/causal response. IIR filters, which are causal, can perfectly be linear in its behaviour. Linear refers usually to amplitude linearity, whilst the difference between IIR and FIR is mostly on the phase/time response.

I think that our ear, as a mechanical semi-linear minimum phase (IIR-style) device, won't differ much in its response with transients or with continuous signals, but I don't know enough about this to talk with authority. That's why I set up the lowpass audibility blind tests with impulsive high-frecuency contents inside a musical signal: http://www.hydrogenaudio.org/forums/index....4894&hl=lowpass

About Frank proposal, I think that as a first test, 96/24 compared to 96/24 -> 44.1/16 -> 96/24 would be enough.

That's why I set up a similar test, but with a digital lowpass and critical musical signals, at the link at my previous post. Only one person I know of has proved to hear the 20 KHz lowpass, none a 21 KHz lowpass.

This is completely off-topic, but I just wanted to say that I always feel enlightened when the major names around HA come out for a high-signal-to-noise discussion. This is one of those times. Thanks, guys.

I just wanted to add something to my paleonthological (?)/biological theory. We are mostly made of water, which conduct ultrasonic waves very well. So there is a chance that we would be responsive to these even though it doesn't come direcly from the ear. The same way as we "feel" a low bass sound instead of hearing it.

So that would mean a headphone is probably not enough to conduct this test.

Easily cheched : just to compare the hearing ability of a fish with the human one. But whatever result is got, it won't make human hear above 20 kHz...

But you're right about ultrasounds in the body. According to this paper, the human body conducts ultrasounds better than air.

interesting study and paper about difference between DSD & PCM 24 bit/176,4 kHz carried out scientifically with abx/double blind,

here the short English paper pdf :
http://www.hfm-detmold.de/eti/projekte/dip..._paper_6086.pdf

German complete websites with pics etc, partly English citings:
http://www.hfm-detmold.de/eti/projekte/dip...m/xdslindex.htm

German complete Theses as pdf, 99 pages:
http://www.hfm-detmold.de/eti/projekte/dip...rbeit%20neu.pdf

short German pdf with pics:
http://www.hfm-detmold.de/eti/projekte/dip...vergleich_d.pdf

I hope, it wasn't posted before.

145 test subjects, all trained listeners in various ways, could not abx at all, they felt very frustrated finding no differences. Counting all samples, it was ca. 1450 vs. 1450, so pure guessing which sample was dsd and which was pcm 24/176,4

Only 4 guys could find difference under headphone conditions with guessing probability less than 5%.
at stereo mode, nothing at surround.


Find more studies or papers at http://www.hfm-detmold.de/eti/ , select German or English version http://www.hfm-detmold.de/eti/indexen.html and look under Theses.

I remember having read this. It must have been discussed somewhere already.

Their statistical model is flawed.

The statement that 2.76% of the completed tests yelded results within the range of "critical probability" (i.e. less than 5% probability of guesswork) is very funny.

If things went completely according to chance, 5% of the completed tests should have yelded results within the range of critical probability, since by definition, critical probability is 5% !

They give detailed information about the results they got. We should be able to see if some results are really significant. But we need to define what is a success and what is a failure.
Implicitly, their definition of a success is "one test at least got a result equal or better than a given score".
We must then specify a probability threshold according to the tested hypothesis. For example, if the possibility that there is an audible difference between DSD and PCM is high, we can use p<0.05. But if we consider this hypothesis as very unlikely, we have to demand a much lower p, so that the tested hypothesis, though unlikely, is still much more probable than chance.
Then we need to compute the minimum individual score s, so that the probability, for random guessing, that at least one people get a score superior or equal to s is inferior or equal to p.

I did not do this, but one of the 145 tests at least was significant, the one who scored 20/20.
They explain that a small click at the beginning of the playback might explain this success, though this click was not conciously audible.

I wonder why they have chosen 176,4 kHz at 24 bits and not something that has a similar data rate like 16 bits. Garf once linked an interesting paper that compared DSD to PCM 176,4 kHz at 8 bits. The PCM thing still seemed to be superior in terms of SNR and linearity.

Sebi

edit: wording

Probably to stay close to real-world applications.